I want to build a complete machine-learning pipeline that can look at images published in social-media posts and automatically recognise what appears in them. The goal is to move from raw, publicly available posts to a trained, production-ready model that delivers accurate labels and confidence scores. I already have sample links to Instagram and Twitter content and can supply API keys for scraping if you need them; feel free to propose an open-source image dataset to augment the training data. Python is my preferred language and I am comfortable with TensorFlow, Keras, PyTorch or a well-structured scikit-learn solution, so choose whichever framework best suits the task. Key expectations • Acquire or curate a balanced set of social-media images (with clear usage rights) and organise it with train/validation/test splits. • Handle preprocessing—downloading, resizing, normalising and, where helpful, augmenting each image. • Design, train and fine-tune a convolutional neural network (transfer learning is fine) aimed at high top-1 accuracy on the validation set. • Provide a concise Jupyter Notebook (or .py script) that reproduces the entire workflow, from data loading to final evaluation metrics. • Save the final model weights and demonstrate inference on unseen social-media posts. • Supply a short read-me explaining setup steps and how I can extend or deploy the model. Acceptance criteria • At least 85 % validation accuracy across the target classes we agree on. • Code runs end-to-end on a fresh environment using the instructions provided. • Clear, inline comments that make the logic easy to follow. If you have past image-recognition work—especially on noisy, real-world data—let me know, and include any ideas you have for improving robustness on constantly evolving social-media content.